In this first episode of 2025, Vinay and Sanjeev Mohan, former Gartner analyst and principal analyst at SanjMo, meet for their annual discussion on the state of the cloud, exploring the transformative power of AI, the shifting dynamics of cloud repatriation, and the resurgence of traditional database technologies like PostgreSQL. They also discuss the implications of AI-driven innovations, such as DeepSeek, and how they are quickly reshaping the competitive landscape. Listen as they dive into the state of the cloud in 2025.

Key insights

⚡AI at Warp Speed: The Race for Efficiency and Accessibility

AI is advancing faster than ever, with models like DeepSeek demonstrating that high-performance AI can be achieved at a fraction of the cost. Sanjeev highlights how AI is transitioning from being a tool for prediction and summarization to powering autonomous agents that can execute tasks on behalf of users. These emerging models and workloads are driving massive investments in AI infrastructure, with companies like Google, Microsoft, and AWS pouring billions into data centers to support the growing demand for AI workloads.

⚡Cloud Repatriation: A Shift in Strategy

While cloud adoption continues to grow, there’s a noticeable trend toward repatriation—moving workloads back on-premises. Sanjeev explains that this shift is driven by the availability of cloud-like features, such as elastic scaling and pay-as-you-go models, now being offered on-premises. However, cloud is here to stay and the future lies in hybrid multi-cloud implementations.

⚡The Rise of PostgreSQL: A Multimodal Powerhouse

PostgreSQL has emerged as a breakout star in the database world, thanks to its ability to handle multiple data formats. Sanjeev predicts that specialized databases will continue to lose ground to multimodal databases like PostgreSQL, which offer flexibility without the added complexity of operating multiple database systems — organizations will continue to consolidate their data stacks.

⚡AI and the Job Market: Disruption and Opportunity

AI is set to disrupt every job, but history shows that new technologies create new opportunities. Sanjeev draws parallels to the introduction of ATMs, which reduced the need for bank tellers, but created space for advisors. Similarly, AI will take over repetitive tasks, allowing humans to focus on more strategic and creative roles. However, the transition will require upskilling and workforce adaptability.

⚡DeepSeek and the Global AI Race

The emergence of DeepSeek, a cost-effective AI model from China, has shaken the global AI landscape. Sanjeev discusses how DeepSeek’s success has forced Western tech giants to rethink their strategies, particularly in light of rising costs and regulatory challenges. The competition between US and Chinese AI models is heating up, with implications for innovation, data privacy, and global AI dominance.

Episode highlights

💡 AI’s Transformative Power: From Prediction to Action [1:53 – 2:44]

Vinay and Sanjeev discuss how AI is evolving from a tool for prediction and summarization to one powering autonomous agents. They explore the implications of this shift for businesses and the growing demand for AI infrastructure.

💡Cloud Repatriation: Why Companies Are Moving Back On-Premises [11:54 – 12:34]

Sanjeev explains the reasons behind the growing trend of cloud repatriation, including the availability of cloud-like features on-premises and the rising costs of cloud services. He also highlights the importance of hybrid multi-cloud models in the future.

💡 PostgreSQL: The Multimodal Database of the Future [28:07 – 28:53]

Vinay and Sanjeev delve into the resurgence of PostgreSQL and its ability to handle multiple data types, making it a strong contender against specialized databases. They discuss how PostgreSQL’s flexibility and scalability are driving its adoption.

💡 DeepSeek: A Game-Changer in the AI Landscape [29:43 – 32:02]

The emergence of DeepSeek, a cost-effective AI model from China, has shaken the global AI landscape. Sanjeev discusses how DeepSeek’s success has forced Western tech giants to rethink their strategies, particularly in light of rising costs and regulatory challenges.

💡 AI and the Job Market: Disruption and Opportunity [35:52 – 37:49]

Sanjeev shares his thoughts on how AI will disrupt the job market, emphasizing the need for upskilling and adaptability. He draws parallels to past technological disruptions, such as the introduction of ATMs, to illustrate how new opportunities emerge.

💡 The Future of AI: Democratization and Accessibility [33:06 – 34:17]

Sanjeev predicts that AI models will become smaller and more accessible, allowing them to run on personal devices like laptops. He discusses how this democratization of AI will drive innovation and adoption across industries.

💡 Recommendations for Navigating the AI Revolution [49:17 – 50:08]

Wrapping up, Sanjeev offers key recommendations for organizations looking to navigate the AI revolution. He emphasizes the importance of data quality, hybrid cloud strategies, and staying adaptable in the face of rapid technological change.

Transcript

Intro: The new numbers, by the way, that are just coming out as we are recording this, show that cloud growth has slowed down. By the way, it is still growing, but it is not growing at the higher percentages it was in the past. So that’s what I’m starting to see.

This is Sovereign DBaaS decoded, a podcast for IT leaders and implementers looking to reliably scale database ops while maintaining control of their data stack. In each episode, we join industry experts to discuss the what and why of sovereignty and how you can implement the sovereign DBaaS concept of your own using open source databases. Deployment models, and tooling. Let’s get started.

Vinay: Hello and welcome to 2025’s first episode of sovereign DBaaS Decoded. I’m Vinay Joosery and this episode is brought to you by Severalnines.

And our guest today is Sanjeev Mohan, former analyst at Gartner. And today principal analyst at SanjMo. Right. And, so you’re a former Gardner Research VP. And then you left about, what, a year and a half ago, two years ago is two… No, its 3 and a half years ago now.

So already time flies. By the way, this is our third year. We are doing this, inaugural podcast of the year. So. Yes. Yes, yes. So thanks for joining us. You know, today, Sanjeev, third year, you know, coming. So, what’s new? I mean, you know, you just got back from Davos.

Sanjeev: Yeah. So actually, every year, of course, we are in IT data space. So every, every year there’s something new except in 2025. So what’s new this year is what’s new this week? Because everything is moving at warp speed and AI is actually transforming faster than we, our wildest expectations I feel this year.

So there’s, So that’s what’s happening. You know, there’s a lot of experimentation. People are starting to use AI or generative AI, not just to predict the next token, the next word, and summarize and write me this and write me that, but more as an assistant, as an agent, taking some autonomous actions on our behalf, even though it’s still early days for agents. But the way the speed at which reasoning models are moving and the speed at which the cost is coming down is a harbinger to how deeply entrenched AI is going to be this year. So, I mean, just literally kicking it off in 2025. Excellent.

Vinay: But hey, again, thank you for coming back to our, let’s say, annual chat. Right. So you work with a good cross-section of the data industry. And that’s a privilege to have you here. So, you know, so to have you here, thank you for sharing your insights.

So what can we expect for today? We’ll get an update from Sanjeev on the state of the cloud. We recap the cloud in the DB space in 2024, and the major influences, events, surprises, and we’ll see what’s going on today and look forward to 2025. But before we do, let’s start with some predictions that we made last year. All right. The first was that the market would shift back to investment mode right now. Has it shifted, or has the bulk of the investment mainly gone to AI? Right. Which was what happened the previous year. What do you say?

Sanjeev: So I think investments are huge. I just—to give an example, as we have regarding this, Google Cloud just announced their earnings for the year, and they showed record revenue growth, but the stock price went down because they’re expecting $75 billion or something like that of investment CapEx in data centers. This was more than what the market expected.

And so the stock went down, which, you know, to some extent is a knee jerk reaction from the market because, we—by, you know, I said earlier these recent models that are, like, getting cheaper. But what’s going to happen is that AI’s going to get so, so much cheaper that organizations are going to have hundreds of agents. So the volume is going to go up. And to serve that volume, you need that investment. So all the big companies—AWS, Microsoft, Google—are still investing vertically more than they have in the past. So I still see investments happening tremendously.

Yeah, and I guess many of these investments are, you know, in the AI kind of sphere. Right, yeah, correct, yeah. And you actually need data centers to serve the AI. Yes, yes, yes. And we should not be surprised. I mean, it seems to be the new cloud for investors, right? A few years back, companies would be pressured to go to the cloud, and, you know, why keep investors happy, right? Why invest CapEx where you can just rent? Right. And it seems to be the same for AI, right? Pretty much every executive team out there is trying to ride the AI wave, yes, even if perhaps they have no real plan for it. And we can see that it’s very tightly coupled with company valuation, right.

Vinay: So we also had some open questions around regulatory, economic, and technological environments, right? So let’s see how they were answered. In terms of regulatory environment, we did not really see, let’s say, more privacy regulation in new countries—maybe because most countries now have their own privacy regulation on the books, right, with provisions coming into force 2024. But there was a key piece of legislation passed on the EU AI act, which wasn’t without its controversy. I don’t know if you have any insights on that, coming off Davos.

Sanjeev: Yeah. So in Davos, it was interesting. I attended a lot of sessions. One session, this was primarily Europeans. If I may, you heard me rant before on your podcast, so here we go, okay? A lot of the Europeans, they were like, “Turn off, GPS, just don’t get on the internet, the big tech companies are stalking you and abusing all the software, all the sensitive information they gather about you.’ So I heard there were a lot of concerns, but then there’s also concern that AI—in fact, in the US now, we have a new president who’s withdrawn the executive order on AI that the previous administration had.

So there’s this big concern that AI is unfettered and can do anything harmful, that there are no controls in place. I am a bit conflicted on this because, first of all, I really see that AI is in such a flux and changing so fast that trying to regulate it isn’t even possible. And those countries that try to regulate it will pay the price of falling behind in innovation. I think I’ve said that before on your podcast. That’s one concern I have. But I also want to say there are already enough guardrails in place, like GDPR, that protect your personal sensitive data; it’s not like AI is going to devalue everything you’ve done and be extremely harmful to humankind. I don’t see that happening. So at some point, we will have more regulation, but I don’t think we’re ready to regulate yet. I’m okay with the regulation we have at this point in time, though my point of view is a bit American-slanted because in America we don’t really take these regulations that seriously, like Europeans do.

But, for example, some of the new models that have just now come out, from either the US or from China, aren’t available in Italy. Why? Because they don’t trust it. So my friends in Europe just VPN and get access to the same content. I give you a dear example of how fear shapes and drives decisions, but in the end, it’ll work out differently. Last week, DeepSeek came out from China—DeepSeek R1 from DeemSeek’s been around at least six months—Microsoft, OpenAI said they’re going to investigate inappropriateness of how the data was trained, and so on. Then literally within two days, Microsoft Azure said, ‘Now we support DeepSeek, all you users who want to use DeepSeek can use it.’ So they went from very anti-DeepSeek to supporting it in two days. That’s our reality now. Everybody supports it. Nvidia has DeepSeek, AWS is full into DeepSeek, we had a whole demo on how to use DeepSeek plus Microsoft Azure, Google Cloud, Snowflake, Databricks. You can use any platform, you can use these models. Anyway, sorry for the long-winded answer, but the upshot is the regulations haven’t been that effective yet in AI. People are more interested in experimenting and trying new technologies and less concerned about regulation.

Vinay: Right, so that was regulation. And yeah, they say America innovates and, you know, Europe regulates, right? So there’s no change there. But in terms of the economic environment, the large tech companies grew their valuations right after making very, very large investments into AI, right? And cloud services still showed strong growth. What do you say?

Sanjeev: Yeah, the move to the cloud is still very strong. Although we’ve talked in the past about repatriation to on-premises—actually, that became quite a big topic in 2024. People talked about it a lot, and I think there was an increase in repatriation from what I saw and heard. But if we take things in relation to each other, cloud migration was still by far much higher than on-prem repatriation. The new numbers, by the way, that are just coming out as we record this, show that cloud growth has slowed down. It’s still growing, but not at the higher percentages it was in the past. So that’s what I’m starting to see.

Vinay: Yeah, growth, but slower. And then in terms of the technology environment, AI still seems very popular—obviously lots of investments—but I wonder about the use cases, because people are still trying to understand the fundamental value of AI for their business beyond just productivity gains across organizational functions, so to speak. And maybe we haven’t really seen the massive use cases that make a real difference?

Sanjeev: Yeah, good point. Again, so much is happening, so much is changing that even my views change day to day. For example, Salesforce at the end of 2024 basically said, ‘No, no, we’re not going to hire a thousand salespeople to sell agents; that number is 2000.’ They doubled the number, came up with SDR agents because that’s what Salesforce does—it’s a CRM system. Now I’m hearing some say, ‘Generic SDR isn’t good, you need personalization.’ So was that a good move or not? We’ll see.

But going back to your point, we’re still looking for that killer use case. I’ve read that AI has led to some major breakthroughs in the medical field—new cures, new diagnoses—but we don’t hear about it mainstream because, has it cured cancer? No, not yet, but it’s a journey. Could I put an AI agent to autonomously run my very complex workflow? Probably not. But could I use it for competitive intelligence? Like if your company wants to know who the new MySQL Sovereign DBaaS companies are, you could have an agent that scrapes the web every day, finds them, summarizes them, and the next morning you come in and see a neat table. That’s obviously a big timesaver. Scraping itself existed, but summarizing and getting smart about how to present data—AI is perfect for that. It makes people more productive.

I published a podcast with Google Cloud leaders a couple days ago, asked them what percentage of code is being written by AI, they said 25%. So we do have these use cases that aren’t mission-critical but are performing well. For me personally, I write a lot of content about what’s coming, what’s new, so I can’t have AI do it because I’m pushing the envelope—AI doesn’t know it yet. But I can have AI polish it. At Gartner, I had a team of world-class editors; now I’m independent and don’t have that, so ChatGPT or Gemini are great editors. Poor actual human editors, right? That’s a whole topic about jobs, though—maybe we won’t get into it deeply.

Vinay: Exactly—jobs is a big topic indeed. But looking at regulation, economy, technology, everything leads back to AI. It feels like 2024 was the year of AI, right? If you weren’t talking about AI, you weren’t having a conversation. So looking back at some significant influencers, events, and surprises of 24: Macroeconomically, we saw inflation stabilizing, interest rates going down in the EU, and yet it seems like everywhere except the US has struggling economies. Geopolitically, wars rage in different regions, Amazon launched a new AI chip, Broadcom ended VMware’s partner program which upset many. That gave renewed energy to Nutanix or OpenStack, CloudStack. RedHat pushed aggressively on KubeVirt. Geico canceled its whole cloud journey after ten years and moved back. Meanwhile, in database technology, we had a vector database gold rush, we even had Stephen Batifol from Milvis on the podcast, but then established vendors integrated vector capability directly into existing products. So how do you think these impacted the market, and did any surprise you?

Sanjeev: I was not surprised. From day one, I said vector databases are not really their own category but more a data type. Storing vector embeddings of unstructured data is basically a vector data type. Having vector indexing and search built into an existing database is the right approach—MySQL, Postgres with pgvector, etc. So to me, vector capability is just a capability, not standalone. I don’t just want to search a PDF, I also want to know who wrote it, which client it came from, the client’s history, etc., all of which lives in a relational or document database. So vector should be closely tied with that data. I shouldn’t have a separate database.

Also in 2024, customers gave up on the so-called modern data stack, because no one wants 15 or 20 different systems to stitch together, each changing schemas. You have multiple data-ingestion products, transformation, catalog, observability—so many categories, each with dozens of startups. The big winners were the cloud hyperscalers, which said, ‘We’ll give you an integrated stack that’s maybe not best in class for each piece, but you get 80% in one place. Your data’s in one location, multiple compute engines on top (Spark, Pandas, Starburst, DuckDB, etc.). Through table formats like Iceberg or Delta or Hudi, you can do that. Meanwhile, we handle the catalog, monitoring, ML tooling, etc.’ That approach led Microsoft to release Fabric, Google to release Dataplex, AWS to release new stuff around S3 tables and SageMaker Lakehouse. So even AWS, which used to pride itself on 16 different databases, consolidated them into a single pane of glass for analytics and machine learning. That’s how people started consolidating data, AI analytics, etc. into one experience.

Vinay: Right. Polyglot persistence was popular, but multimodal from an existing vendor can reduce complexity. Some enterprises still want specialized data stores for each job, some say ‘Now MariaDB does vector, we’re good!’ So yeah, there’s a blend. I’ve heard MariaDB’s vector implementation is good, maybe even fastest. You’d know more. Meanwhile, Postgres usage exploded, so pgvector has a massive footprint, and Postgres was arguably the breakout star of 2024. Let’s shift gears to today. Obviously, AI remains marquee. Let’s start with DeepSeek surfacing, raising questions about the US industry’s strategy of throwing money at AI, plus the government’s strategy of limiting China’s access to the latest chips. You mentioned big tech was initially skeptical of DeepSeek, then everyone jumped in. Thoughts?

Sanjeev: Yeah, so there’s a huge Cold War of competition. Initially, folks question transparency from Chinese companies, plus IP violations in the past. But it was a wake-up call: Meta said Llama4 might cost a billion dollars to train, and you can’t sustain spending billions on each iteration. Then along comes DeepSeek claiming it cost a fraction. They published an open paper. After a few days of doubt, better logic prevailed—turns out it’s a good model.

People realized you can locally host a fork of DeepSeek and pay basically zero. So that eliminates concerns about data going to China. You get a state-of-the-art reasoning model for near-zero cost. Some argue censorship—like if you ask it about Chinese leadership or Tiananmen, you get nothing—but devs don’t care, they just want a capable foundation model for tasks. So after two days of investigating, Microsoft, Nvidia, AWS, etc. said, ‘We support DeepSeek.’ Suddenly it’s ubiquitous. But by the time we talk, it might be history because 15 other Chinese companies are competing, Alibaba has 2.5, etc. The sad part is that intense competition is happening in China, not the US, but the US will have to adapt. Meanwhile, we’re on the cusp of every model soon being small enough to run on a laptop, no massive GPU farm needed. That’s how quickly it’s moving.

Vinay: Satya Nadella tweeted about Jevons paradox—AI gets more efficient and accessible, usage skyrockets, unstoppable growth. It’s democratizing the resource, costs drop, demand rises. Maybe we’re at peak hype cycle? Or maybe not?

Sanjeev: I don’t think we’re at peak. Hype’s going higher. Costs are plummeting. It’s still changing so fast. One question is job market disruption—every job from yours to mine might be touched by AI. But I avoid that deeper discussion; historically, new technologies replaced busywork. We used to fear computers. The 1960s said we’d have endless leisure time. That didn’t happen. So I think we’ll keep working, but do higher-level tasks while AI does the grunt stuff. Like when ATMs took over dispensing cash—tellers became private bankers or financial advisors. Similarly, DBAs used to tune hundreds of Oracle parameters; now we have self-tuning DBs, so DBAs do more strategic tasks tied to business. That alignment with business is how you stay relevant.

Vinay: Great example. So, turning to more pedestrian concerns: CEO surveys say confidence has turned a corner for 2025, renewed appetite for investment. Might we see room for classic operational data stores if everything’s overshadowed by AI? Possibly the ‘boring stuff’ like stable ops?

Sanjeev: Yes, absolutely. AI’s limited by data quality, accessibility, governance, security, so you can’t have great AI without great operational data. So ironically, the old fundamentals matter more than ever—fix data or forget AI. Might be operational or synthetic or unstructured. The volume is exploding, but we still need the basics. So ironically, that’s fueling the ‘boring’ data management tasks.

Vinay: Right, and you mentioned cloud is slowing a bit. Maybe repatriation is accelerating. Does that mean the cloud is shifting from a destination to an operating model?

Sanjeev: It already has. One reason for repatriation is on-prem can now deliver the elasticity, pay-as-you-go, ephemeral approach we expect from the cloud. But mostly, I see hybrid multi-cloud. Clients say, ‘Leave data on-prem, bring compute to it seamlessly.’ That’s why cloud providers push Kubernetes on-prem, containerizing BigQuery Omni or similar. So you can run a cloud analytics engine locally but still pay the vendor. It’s beneficial for them and for customers who can’t move data out.

Vinay: Right. Some can’t even power all their on-prem hardware, so it’s tricky. But from a CIO perspective, to dabble in AI, maybe it’s easier to rent from Amazon or Microsoft for six months rather than buy million-dollar GPUs. So do your experimentation in the cloud, maybe repatriate later once you have the plan.

Sanjeev: Correct—experimentation is trivial in the cloud, just a credit card. On-prem you have to size, order, provision, stack, rack. By the time it’s in place you might miss the bus or it’s outdated. So experiment in the cloud. If you have a giant backlog of unstructured data, see if there’s ROI. People do repatriation because cloud can be expensive if you fail to configure it properly. Lift-and-shift with your old habits, you’ll see the bill blow up daily. Simple stuff like developer logs that were free on-prem now go to CloudWatch and cost an arm and a leg. So folks say, ‘Cloud is too expensive!’ but it’s usually because they used it ineffectively.

Vinay: Yep, that makes sense. So looking forward to ’25: continued AI investments, cost of AI infra slashed, maybe more cloud repatriation. What about the global economy with trade tariffs, etc.? Do you agree with CEOs who say 2025 will be a great year for business?

Sanjeev: I’m guarded. One of the ML companies I work with just had a massive layoff yesterday. There’s a sense the S&P 500 is up, but it’s really the S&P 7 driving it; the other 493 are struggling. I’m optimistic, but many people I know have been job-hunting a year. Meanwhile, the US is doing great on paper. So I won’t say 2025 is guaranteed fantastic. Who knows?

Vinay: Sure, the EU has lagged, Mario Draghi said we need €800B a year in investments. We see some new HPC/hyperscaler efforts in Europe, like Evroc. Everyone’s chasing AI for valuation. So is that a bubble? Will it burst? Will we see results in 25 in actual revenue growth? We’ll see. But yes, it’s reminiscent of the dotcom era—some companies will fail, the tech will remain. Possibly the same for AI. We might see that pattern.

Sanjeev: Exactly. Dotcom boom and bust was bright for a year, then crashed. But the internet didn’t die; it just kept evolving. WebVan failed, but now grocery delivery is commonplace. Even if AI companies fail, AI’s unstoppable. Maybe not 2025, maybe 2028, but it’ll be embedded everywhere eventually.

Vinay: Right. So final question: what does all this mean for databases? Any 2025 predictions?

Sanjeev: Data has never been more important. Mega trend of all time is data. Database vendors will thrive. Some folks want specialized databases, but many specialized APIs are appearing in multimodal DBs. Postgres can store JSON, vector, etc. So do you need a separate NoSQL? Maybe not unless it’s Mongo. Same for graph—Neo4j leads, but many do a bit of graph inside a bigger DB. So the database market’s in good shape. MongoDB stock might be down, Oracle’s is up because of AI. We’ll see, but databases are critical.

Vinay: Got it. Well, that’s a wrap. Thank you, Sanjeev—always great having you. We looked at 2024’s stabilizing economy, AI dominating mindshare, repatriation, some big surprises. We’ll see how 2025 goes, watch for more Geico-like moves. Thanks for listening, and have a great 2025.

Sanjeev: Thank you.